Alignment and Uniformity¶
In one minute
A good representation keeps similar instances close together and separates different ones. Wang and Isola (2020) call these two properties alignment and uniformity and describe both on a hypersphere.
A good representation should be able to
- separate different instances, and
- cluster similar instances.
Wang et al proposed two concepts that matches the above two ideas, alignment and uniformity, on a hypersphere1.

From Wang et al1.
Related: Contrastive Predictive Coding · Deep Infomax
-
Wang T, Isola P. Understanding Contrastive Representation Learning through Alignment and Uniformity on the Hypersphere. arXiv [cs.LG]. 2020. Available: http://arxiv.org/abs/2005.10242 ↩↩